Methods
We conducted a secondary analysis of body temperature by using data from the National Heart, Lung and Blood Institute (NHLBI) ARDS Network Fluid and Catheter Treatment Trial. This multicenter factorial study randomized patients with acute lung injury for 48 hours or less to receive a central venous catheter or a pulmonary artery catheter and to receive either liberal or conservative fluid management strategies per protocol. The institutional review boards of participating centers and the NHLBI approved the original study. Written consent was obtained from the patient participants or their legal surrogates in the original study. Certification from the institutional review board at the investigators' center was obtained for this secondary analysis.
Adult patients who met the American-European Consensus criteria for acute lung injury for 48 hours or less were eligible for study enrollment. With the exception of 0.2% of this study's sample, patients met the recently published criteria for the Berlin definition of ARDS. Exclusion criteria included presence of ARDS for more than 48 hours, presence of a pulmonary artery catheter before study enrollment, presence of chronic conditions that could influence compliance with the study protocol or ventilator weaning, and terminal conditions with estimated 6-month mortality of greater than 50%. Because of missing data on body temperature and score on the Acute Physiology and Chronic Health Evaluation (APACHE) III, 31 patients were excluded from the original sample of 1000 patients.
Measurement of Variables
The sources of baseline measurements of body temperature in the original study included rectal, tympanic, and axillary sites. Baseline temperature was obtained from the 4-hour period preceding randomization, which occurred immediately after consent was obtained. Body temperature was measured at the same time each day from rectal, tympanic, axillary, or pulmonary artery catheter sites and recorded for up to 7 days. Temperature ranges used to create 5 groups were selected on the basis of definitions of moderate to deep hypothermia (< 34°C), mild hypothermia (34°C-35.9°C), normothermia (36°C-38.2°C), fever (38.3°C-39.4°C), and high fever (≥39.5°C).
Patients were followed up for 90 days after study enrollment or until death, whichever occurred first. The APACHE III score was calculated from patients' baseline data. One of the following causes of primary lung injury was selected for each patient: trauma, sepsis, multiple transfusions, pneumonia, aspiration, or other causes.
Statistical Analysis
An independent-samples t test was conducted to compare baseline temperatures for survivors and nonsurvivors. In order to control for potential confounding variables, multiple logistic regression was performed to assess the impact of 3 factors on the likelihood of mortality at 90 days in patients with ARDS. The 3 factors in the model were baseline temperature, primary cause of ARDS, and severity of illness, measured by the APACHE III score. These variables were included because of their potential physiological and clinical significance as well as their significant association with mortality in univariate analyses. In addition, as a sensitivity analysis to explore whether hypothermia influenced the results of the study, the multiple logistic regression was repeated with exclusion of patients with body temperatures less than 36°C. Multiple logistic regression was also repeated using temperature from day 2 of the study in place of baseline temperature to determine whether the relationship was sustained at another time point early in the ARDS trajectory.
To better understand the relationship between body temperature and mortality, we used 5 categories of baseline temperature (moderate to deep hypothermia, mild hypothermia, normothermia, fever, and high fever) and used logistic regression to test for a trend in the mortality among the temperature groups. Baseline characteristics were compared among the 5 temperature groups by using 1-way analysis of variance for continuous variables and χ analysis for categorical variables.
Because temperature is part of the APACHE III score calculation, correlation analyses and collinearity diagnostics of the independent variables were completed, and low correlations ruled out concern about multicollinearity issues. The Hosmer-Lemeshow test was used to assess the goodness of fit of the model. Odds ratios and 95% confidence intervals were calculated. Statistical tests were 2-sided and differences were considered significant at P less than .05. Data were analyzed with SPSS computer software, version 21 (SPSS, Inc).